Autonomous Research Agents
Overview
Autonomous research agents use language models to plan and carry out research tasks, from literature search and hypothesis generation to coding, experiments, analysis, and reporting. Common approaches include tool-using coding agents, teams of specialized agents, persistent memory, and training with supervised or reinforcement learning; execution harnesses and formal or domain-specific checks help constrain their work. Agents can perform well on bounded, executable tasks such as model optimization, code reproduction, and formal proof search, but open-ended research remains harder, especially when novelty or scientific validity is difficult to measure. A central challenge is ensuring that reported claims follow from reliable experiments and evidence, which motivates independent verification, held-out evaluation, provenance tracking, and continued human oversight.
Latest papers
324 papers60 papers in the last four weeks, up 62% on the four weeks before. 0.3% of all new papers.